In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!
Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the Jupyter Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.
The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this Jupyter notebook.
In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!
We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.
Make sure that you've downloaded the required human and dog datasets:
Download the dog dataset. Unzip the folder and place it in this project's home directory, at the location /dog_images.
Download the human dataset. Unzip the folder and place it in the home directory, at location /lfw.
Note: If you are using a Windows machine, you are encouraged to use 7zip to extract the folder.
In the code cell below, we save the file paths for both the human (LFW) dataset and dog dataset in the numpy arrays human_files and dog_files.
import numpy as np
from glob import glob
# load filenames for human and dog images
human_files = np.array(glob("/data/lfw/*/*"))
dog_files = np.array(glob("/data/dog_images/*/*/*"))
# print number of images in each dataset
print('There are %d total human images.' % len(human_files))
print('There are %d total dog images.' % len(dog_files))
In this section, we use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images.
OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory. In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.
import cv2
import matplotlib.pyplot as plt
%matplotlib inline
# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')
# load color (BGR) image
img = cv2.imread(human_files[0])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# find faces in image
faces = face_cascade.detectMultiScale(gray)
# print number of faces detected in the image
print('Number of faces detected:', len(faces))
# get bounding box for each detected face
for (x,y,w,h) in faces:
# add bounding box to color image
cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.
In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.
We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
img = cv2.imread(img_path)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray)
return len(faces) > 0
Question 1: Use the code cell below to test the performance of the face_detector function.
human_files have a detected human face? dog_files have a detected human face? Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.
Answer: (You can print out your results and/or write your percentages in this cell)
Percentage of the first 100 images in human_files with a detected human face: 98.0%
Percentage of the first 100 images in dog_files with a detected human face: 17.0%
from tqdm import tqdm
human_files_short = human_files[:100]
dog_files_short = dog_files[:100]
count_faces=0
count_dogs=0
#-#-# Do NOT modify the code above this line. #-#-#
## TODO: Test the performance of the face_detector algorithm
## on the images in human_files_short and dog_files_short.
for i in range(len(human_files_short)):
count_faces+=face_detector(human_files_short[i])
print("Percentage of the first 100 images in human_files have a detected human face: {}%".format(count_faces/len(human_files_short)*100))
for i in range(len(dog_files_short)):
count_dogs+=face_detector(dog_files_short[i])
print("Percentage of the first 100 images in dog_files have a detected human face: {}%".format(count_dogs/len(dog_files_short)*100))
We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on human_files_short and dog_files_short.
### (Optional)
### TODO: Test performance of anotherface detection algorithm.
### Feel free to use as many code cells as needed.
In this section, we use a pre-trained model to detect dogs in images.
The code cell below downloads the VGG-16 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories.
import torch
import torchvision.models as models
# define VGG16 model
VGG16 = models.vgg16(pretrained=True)
# check if CUDA is available
use_cuda = torch.cuda.is_available()
# move model to GPU if CUDA is available
if use_cuda:
VGG16 = VGG16.cuda()
Given an image, this pre-trained VGG-16 model returns a prediction (derived from the 1000 possible categories in ImageNet) for the object that is contained in the image.
In the next code cell, you will write a function that accepts a path to an image (such as 'dogImages/train/001.Affenpinscher/Affenpinscher_00001.jpg') as input and returns the index corresponding to the ImageNet class that is predicted by the pre-trained VGG-16 model. The output should always be an integer between 0 and 999, inclusive.
Before writing the function, make sure that you take the time to learn how to appropriately pre-process tensors for pre-trained models in the PyTorch documentation.
from PIL import Image
import torchvision.transforms as transforms
def VGG16_predict(img_path):
'''
Use pre-trained VGG-16 model to obtain index corresponding to
predicted ImageNet class for image at specified path
Args:
img_path: path to an image
Returns:
Index corresponding to VGG-16 model's prediction
'''
## TODO: Complete the function.
## Load and pre-process an image from the given img_path
## Return the *index* of the predicted class for that image
image = Image.open(img_path)
in_transform = transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406),
(0.229, 0.224, 0.225))])
image = in_transform(image)
# add batch dimension
image = image.unsqueeze_(0)
VGG16.eval()
output = VGG16(image)
#print(output.shape)
#predicted = torch.argmax(output)
index = output.data.numpy().argmax()
return index # predicted class index
# Test the function
VGG16_predict(dog_files[8000])
While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained VGG-16 model, we need only check if the pre-trained model predicts an index between 151 and 268 (inclusive).
Use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
## TODO: Complete the function.
prediction = VGG16_predict(img_path)
if prediction > 150 and prediction < 269:
output = True
else:
output = False
return output # true/false
#test the function
print(dog_detector(dog_files[8000]))
print(dog_detector(human_files[8000]))
Question 2: Use the code cell below to test the performance of your dog_detector function.
human_files_short have a detected dog? dog_files_short have a detected dog?Answer:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
human_files_short = human_files[:100]
dog_files_short = dog_files[:100]
count_human_dog=0
count_dog_dog=0
#-#-# Do NOT modify the code above this line. #-#-#
## TODO: Test the performance of the face_detector algorithm
## on the images in human_files_short and dog_files_short.
print(VGG16)
for i in range(len(human_files_short)):
count_human_dog+=dog_detector(human_files_short[i])
print("Percentage of the first 100 images in human_files have a detected dog: {}%".format(count_human_dog/len(human_files_short)*100))
for i in range(len(dog_files_short)):
count_dog_dog+=dog_detector(dog_files_short[i])
print("Percentage of the first 100 images in dog_files have a detected dog: {}%".format(count_dog_dog/len(dog_files_short)*100))
We suggest VGG-16 as a potential network to detect dog images in your algorithm, but you are free to explore other pre-trained networks (such as Inception-v3, ResNet-50, etc). Please use the code cell below to test other pre-trained PyTorch models. If you decide to pursue this optional task, report performance on human_files_short and dog_files_short.
### (Optional)
### TODO: Report the performance of another pre-trained network.
### Feel free to use as many code cells as needed.
import torch
import torchvision.models as models
# define VGG16 model
RESNET = models.resnet18(pretrained=True)
# check if CUDA is available
use_cuda = torch.cuda.is_available()
# move model to GPU if CUDA is available
if use_cuda:
RESNET = RESNET.cuda()
from PIL import Image
import torchvision.transforms as transforms
def RESNET_predict(img_path):
'''
Use pre-trained VGG-16 model to obtain index corresponding to
predicted ImageNet class for image at specified path
Args:
img_path: path to an image
Returns:
Index corresponding to VGG-16 model's prediction
'''
## TODO: Complete the function.
## Load and pre-process an image from the given img_path
## Return the *index* of the predicted class for that image
image = Image.open(img_path)
in_transform = transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406),
(0.229, 0.224, 0.225))])
image = in_transform(image)
# add batch dimension
image = image.unsqueeze_(0)
RESNET.eval()
output = RESNET(image)
#print(output.shape)
#predicted = torch.argmax(output)
index = output.data.numpy().argmax()
return index # predicted class index
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
## TODO: Complete the function.
prediction = RESNET_predict(img_path)
if prediction > 150 and prediction < 269:
output = True
else:
output = False
return output # true/false
human_files_short = human_files[:100]
dog_files_short = dog_files[:100]
count_human_dog=0
count_dog_dog=0
#-#-# Do NOT modify the code above this line. #-#-#
## TODO: Test the performance of the face_detector algorithm
## on the images in human_files_short and dog_files_short.
print("RESNET")
for i in range(len(human_files_short)):
count_human_dog+=dog_detector(human_files_short[i])
print("Percentage of the first 100 images in human_files have a detected dog: {}%".format(count_human_dog/len(human_files_short)*100))
for i in range(len(dog_files_short)):
count_dog_dog+=dog_detector(dog_files_short[i])
print("Percentage of the first 100 images in dog_files have a detected dog: {}%".format(count_dog_dog/len(dog_files_short)*100))
Observations:
Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 10%. In Step 4 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.
We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have trouble distinguishing between a Brittany and a Welsh Springer Spaniel.
| Brittany | Welsh Springer Spaniel |
|---|---|
![]() |
![]() |
It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).
| Curly-Coated Retriever | American Water Spaniel |
|---|---|
![]() |
![]() |
Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.
| Yellow Labrador | Chocolate Labrador | Black Labrador |
|---|---|---|
![]() |
![]() |
![]() |
We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.
Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!
Use the code cell below to write three separate data loaders for the training, validation, and test datasets of dog images (located at dog_images/train, dog_images/valid, and dog_images/test, respectively). You may find this documentation on custom datasets to be a useful resource. If you are interested in augmenting your training and/or validation data, check out the wide variety of transforms!
import os
import torch
from torchvision import datasets
import torchvision.transforms as transforms
!pip install torchsummary
import torch.nn as nn
import torch.nn.functional as F
from torchsummary import summary
import numpy as np
import matplotlib.pyplot as plt
from glob import glob
import pandas as pd
import torch.optim as optim
use_cuda = torch.cuda.is_available()
### TODO: Write data loaders for training, validation, and test sets
## Specify appropriate transforms, and batch_sizes
categories_train = np.array(glob("/data/dog_images/train/*/*"))
for i in range(len(categories_train)):
categories_train[i] = categories_train[i][23:26]
unique_train, counts_train = np.unique(categories_train, return_counts=True)
categories_test = np.array(glob("/data/dog_images/test/*/*"))
for i in range(len(categories_test)):
categories_test[i] = categories_test[i][22:25]
unique_test, counts_test = np.unique(categories_test, return_counts=True)
categories_valid = np.array(glob("/data/dog_images/valid/*/*"))
for i in range(len(categories_valid)):
categories_valid[i] = categories_valid[i][23:26]
unique_valid, counts_valid = np.unique(categories_valid, return_counts=True)
fig, axes = plt.subplots(nrows=1, ncols=3, sharex=True, sharey=True, figsize=(25,4)) #figsize (width,height) in inches
plt.subplot(1, 3, 1)
plt.bar(unique_train, counts_train)
plt.subplot(1, 3, 2)
plt.bar(unique_test, counts_test)
plt.subplot(1, 3, 3)
plt.bar(unique_valid, counts_valid)
plt.show()
Oberservations:
total = len(categories_train) + len(categories_test) + len(categories_valid)
print("Total data: {}".format(total))
print("Length of train_data: {}, {:.1f}%".format(len(categories_train),len(categories_train)/total*100))
print("Length of test_data: {}, {:.1f}%".format(len(categories_test),len(categories_test)/total*100))
print("Length of validate_data: {}, {:.1f}%".format(len(categories_valid),len(categories_valid)/total*100))
# Take a look at the data
import random
from PIL import Image
import matplotlib.pyplot as plt
import torchvision.transforms as transforms
transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
#transforms.RandomVerticalFlip(),
#transforms.RandomRotation(degrees=15),
transforms.Resize((224,224))
])
dog_files = np.array(glob("/data/dog_images/*/*/*"))
index=[]
for i in range(30):
index.append(random.randint(1,len(dog_files)-1))
print(index)
fig, axes = plt.subplots(nrows=6, ncols=5, sharex=True, sharey=True, figsize=(25,25)) #figsize (width,height) in inches
for i in range(30):
plt.subplot(6, 5, i+1)
plt.imshow(transform(Image.open(dog_files[index[i]])))
plt.xticks([])
plt.yticks([])
plt.show()
Observations:
transform = {
'train' : transforms.Compose([
transforms.RandomHorizontalFlip(), #trasnformations do not increase number of images
transforms.RandomRotation(degrees=15),
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),
'test' : transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),
'valid' : transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
}
# number of subprocesses to use for data loading
num_workers = 0
# how many samples per batch to load
batch_size = 20
files = {
'train' : datasets.ImageFolder(root='/data/dog_images/train', transform=transform['train']),
'test' : datasets.ImageFolder(root='/data/dog_images/test', transform=transform['test']),
'valid' : datasets.ImageFolder(root='/data/dog_images/valid', transform=transform['valid'])
}
loaders_scratch = {
'train' : torch.utils.data.DataLoader(files['train'], batch_size=batch_size, num_workers=num_workers, shuffle=True),
'test' : torch.utils.data.DataLoader(files['test'], batch_size=batch_size, num_workers=num_workers, shuffle=True),
'valid' : torch.utils.data.DataLoader(files['valid'], batch_size=batch_size, num_workers=num_workers, shuffle=True)
}
Question 3: Describe your chosen procedure for preprocessing the data.
Answer: I resized the images by stretching/compressing them to 224 by 224, as it is an easy size to work with (similar to images used to train VGG). The images are at different positions and thus, I did not use central cropping.
To augment the dataset, I added horizontal flips and rotations to better fit the distribution of the image set as I observed that the heads of some dogs are tilted in some images.
Create a CNN to classify dog breed. Use the template in the code cell below.
class Net(nn.Module):
### TODO: choose an architecture, and complete the class
def __init__(self):
super(Net, self).__init__()
## Define layers of a CNN
self.conv1 = nn.Conv2d(3, 64, 3, padding=1)
self.conv2 = nn.Conv2d(64, 128, 3, padding=1)
self.conv3 = nn.Conv2d(128, 256, 3, padding=1)
self.conv4 = nn.Conv2d(256, 512, 3, padding=1)
self.conv4a = nn.Conv2d(512, 512, 3, padding=1)
# max pooling layer
self.pool1 = nn.MaxPool2d(2, 2, ceil_mode=False)
# linear layer
self.fc1 = nn.Linear(512*7*7, 4096)
self.fc2 = nn.Linear(4096, 133)
# dropout layer
self.dropout = nn.Dropout(0.25)
def forward(self, x):
#224-112-56-28-14-7
x = self.conv1(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv2(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv3(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv4(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv4a(x)
x = F.relu(x)
x = self.pool1(x)
# flatten image input
x = x.view(-1, 512 * 7 * 7)
x = self.dropout(x)
x = F.relu(self.fc1(x))
x = self.dropout(x)
x = self.fc2(x)
return x
#-#-# You so NOT have to modify the code below this line. #-#-#
# instantiate the CNN
model_scratch = Net()
#print(summary(model_scratch, (3, 224, 224))) #Does not work with cuda
print(model_scratch)
# move tensors to GPU if CUDA is available
if torch.cuda.is_available():
model_scratch = model_scratch.cuda()
criterion_scratch = nn.CrossEntropyLoss()
optimizer_scratch = optim.SGD(model_scratch.parameters(), lr=0.1)
model_scratch = train(5, loaders_scratch, model_scratch, optimizer_scratch, criterion_scratch, use_cuda, 'model_scratch.pt')
test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
#continue to train the model as it seems to be still improving (epochs 7 to 11 shown below)
model_scratch = train(5, loaders_scratch, model_scratch, optimizer_scratch, criterion_scratch, use_cuda, 'model_scratch.pt')
test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
Potential improvements and trade-offs::
# define the CNN architecture
class Net(nn.Module):
### TODO: choose an architecture, and complete the class
def __init__(self):
super(Net, self).__init__()
## Define layers of a CNN
self.conv1 = nn.Conv2d(3, 16, 3, padding=1)
self.conv2 = nn.Conv2d(16, 32, 3, padding=1)
self.conv3 = nn.Conv2d(32, 64, 3, padding=1)
self.conv4 = nn.Conv2d(64, 128, 3, padding=1)
self.conv5 = nn.Conv2d(128, 256, 3, padding=1)
#batch normalization - after conv later OR after ReLu
self.conv_bn1 = nn.BatchNorm2d(16)
self.conv_bn2 = nn.BatchNorm2d(32)
self.conv_bn3 = nn.BatchNorm2d(64)
self.conv_bn4 = nn.BatchNorm2d(128)
self.conv_bn5 = nn.BatchNorm2d(256)
# max pooling layer
self.pool1 = nn.MaxPool2d(2, 2, ceil_mode=False)
self.pool_global = nn.AvgPool2d(14) #global max pooling reduces model size
# linear layer
#self.fc1 = nn.Linear(256*14*14, 4096)
#self.fc2 = nn.Linear(4096, 133)
self.fc = nn.Linear(256, 133)
# dropout layer
self.dropout = nn.Dropout(0.1)
def forward(self, x):
#224-112-56-28-14-7
x = self.conv1(x)
x = self.conv_bn1(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv2(x)
x = self.conv_bn2(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv3(x)
x = self.conv_bn3(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv4(x)
x = self.conv_bn4(x)
x = F.relu(x)
x = self.pool1(x)
x = self.conv5(x)
x = self.conv_bn5(x) # might not need in the last layer. ref batch norm lesson
x = F.relu(x)
x = self.pool_global(x) #256x7x7
# flatten image input
x = x.view(-1, 256 * 1 * 1)
#x = F.relu(self.fc1(x))
x = self.dropout(x)
#x = F.softmax(self.fc(x))
x = self.fc(x)
return x
#-#-# You so NOT have to modify the code below this line. #-#-#
# instantiate the CNN
model_scratch = Net()
#print(summary(model_scratch, (3, 224, 224))) #Does not work with cuda
print(model_scratch)
# move tensors to GPU if CUDA is available
if torch.cuda.is_available():
model_scratch = model_scratch.cuda()
print(summary(model_scratch, (3, 224, 224)))
criterion_scratch = nn.CrossEntropyLoss()
optimizer_scratch = optim.SGD(model_scratch.parameters(), lr=0.1)
model_scratch = train(10, loaders_scratch, model_scratch, optimizer_scratch, criterion_scratch, use_cuda, 'model_scratch_2.pt')
test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
Further Observations after tweaking model:
Potential model improvements:
Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step.
Answer:
I started with a simplied version of VGG, with 5 convolution layers (3-64-128-256-512-512) followed by ReLu activation and MaxPool. There were 2 last linear layers with a dropout of 0.25 in between. The training was slow and I got an accuracy of about 16%. There were signs of over-fitting around the 8th epoch.
Test Loss: 3.672204 Test Accuracy: 16% (135/836)
After reading some papers, I added batch normalisation in between the Convolution and ReLu functions, and decreased the dropout to 0.15 and replaced the 2 linear layers by a Global Max Pool to reduce the model size and for the model to train more quickly. This model trained more quickly but the acurracy obtained was lower. Overfitting was observed around epoch 7 as training loss decreased while validation loss did not. The best model was obtained after 8 epochs with an accuracy of 10%.
Test Loss: 4.547163 Test Accuracy: 10% (88/836)
I then added a dropout before the last linear layer. 0.2 was too high but 0.1 proved to improve the model slightly.
Test Loss: 3.973795 Test Accuracy: 12% (101/836)
Use the next code cell to specify a loss function and optimizer. Save the chosen loss function as criterion_scratch, and the optimizer as optimizer_scratch below.
import torch.optim as optim
### TODO: select loss function
criterion_scratch = nn.CrossEntropyLoss()
### TODO: select optimizer
optimizer_scratch = optim.SGD(model_scratch.parameters(), lr=0.1)
#exp_lr_scheduler = optim.lr_scheduler.StepLR(optimizer_scratch, step_size=7, gamma=0.1)
#optimizer_scratch = optim.RMSprop(model_scratch.parameters(), lr=0.2)
#optimizer_scratch = optim.Adam(model_scratch.parameters(), lr=0.5)
Train and validate your model in the code cell below. Save the final model parameters at filepath 'model_scratch.pt'.
import torch
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
def train(n_epochs, loaders, model, optimizer, criterion, use_cuda, save_path):
"""returns trained model"""
# initialize tracker for minimum validation loss
valid_loss_min = np.Inf
for epoch in range(1, n_epochs+1):
# initialize variables to monitor training and validation loss
train_loss = 0.0
valid_loss = 0.0
###################
# train the model #
###################
model.train()
for batch_idx, (data, target) in enumerate(loaders['train']):
# move to GPU
if use_cuda:
data, target = data.cuda(), target.cuda()
## find the loss and update the model parameters accordingly
## record the average training loss, using something like
## train_loss = train_loss + ((1 / (batch_idx + 1)) * (loss.data - train_loss))
# clear the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: compute predicted outputs by passing inputs to the model
output = model(data)
# calculate the batch loss
loss = criterion(output, target)
# backward pass: compute gradient of the loss with respect to model parameters
loss.backward()
# perform a single optimization step (parameter update)
optimizer.step()
# update training loss
train_loss += ((1 / (batch_idx + 1)) * (loss.data - train_loss))
######################
# validate the model #
######################
model.eval()
for batch_idx, (data, target) in enumerate(loaders['valid']):
# move to GPU
if use_cuda:
data, target = data.cuda(), target.cuda()
## update the average validation loss
# forward pass: compute predicted outputs by passing inputs to the model
output = model(data)
# calculate the batch loss
loss = criterion(output, target)
# update average validation loss
valid_loss += ((1 / (batch_idx + 1)) * (loss.data - valid_loss))
# print training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}'.format(
epoch,
train_loss,
valid_loss
))
#for name, param in model.fc.named_parameters():
# print(name, param.data)
## TODO: save the model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(model.state_dict(), save_path)
valid_loss_min = valid_loss
# return trained model
return model
# train the model
model_scratch = train(10, loaders_scratch, model_scratch, optimizer_scratch,
criterion_scratch, use_cuda, 'model_scratch.pt')
# load the model that got the best validation accuracy
#model_scratch.load_state_dict(torch.load('model_scratch.pt'))
Try out your model on the test dataset of dog images. Use the code cell below to calculate and print the test loss and accuracy. Ensure that your test accuracy is greater than 10%.
def test(loaders, model, criterion, use_cuda):
# monitor test loss and accuracy
test_loss = 0.
correct = 0.
total = 0.
model.eval()
for batch_idx, (data, target) in enumerate(loaders['test']):
# move to GPU
if use_cuda:
data, target = data.cuda(), target.cuda()
# forward pass: compute predicted outputs by passing inputs to the model
output = model(data)
# calculate the loss
loss = criterion(output, target)
# update average test loss
test_loss = test_loss + ((1 / (batch_idx + 1)) * (loss.data - test_loss))
# convert output probabilities to predicted class
pred = output.data.max(1, keepdim=True)[1]
# compare predictions to true label
correct += np.sum(np.squeeze(pred.eq(target.data.view_as(pred))).cpu().numpy())
total += data.size(0)
print('Test Loss: {:.6f}\n'.format(test_loss))
print('\nTest Accuracy: %2d%% (%2d/%2d)' % (
100. * correct / total, correct, total))
# call test function
#test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.
Use the code cell below to write three separate data loaders for the training, validation, and test datasets of dog images (located at dogImages/train, dogImages/valid, and dogImages/test, respectively).
If you like, you are welcome to use the same data loaders from the previous step, when you created a CNN from scratch.
transform = {
'train' : transforms.Compose([
transforms.RandomHorizontalFlip(), #trasnformations do not increase number of images
transforms.RandomRotation(degrees=15),
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))]),
'test' : transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))]),
'valid' : transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])
}
# number of subprocesses to use for data loading
num_workers = 0
# how many samples per batch to load
batch_size = 20
files = {
'train' : datasets.ImageFolder(root='/data/dog_images/train', transform=transform['train']),
'test' : datasets.ImageFolder(root='/data/dog_images/test', transform=transform['test']),
'valid' : datasets.ImageFolder(root='/data/dog_images/valid', transform=transform['valid'])
}
loaders_transfer = {
'train' : torch.utils.data.DataLoader(files['train'], batch_size=batch_size, num_workers=num_workers, shuffle=True),
'test' : torch.utils.data.DataLoader(files['test'], batch_size=batch_size, num_workers=num_workers, shuffle=True),
'valid' : torch.utils.data.DataLoader(files['valid'], batch_size=batch_size, num_workers=num_workers, shuffle=True)
}
Use transfer learning to create a CNN to classify dog breed. Use the code cell below, and save your initialized model as the variable model_transfer.
import torchvision.models as models
import torch.nn as nn
import torch.optim as optim
## TODO: Specify model architecture
model_transfer = models.vgg16(pretrained=True)
# Freeze training for all "features" layers
for param in model_transfer.parameters():
param.requires_grad = False
#Unfreeze last conv layer -> get better results?
#for param in model_transfer.classifier[3].parameters():
#for param in model_transfer.layer4[1].conv2.parameters():
# param.requires_grad = True
n_inputs = model_transfer.classifier[6].in_features
last_layer = nn.Linear(n_inputs, 133, bias = True)
model_transfer.classifier[6] = last_layer
#for param in model_transfer.classifier[6].parameters():
# param.requires_grad = True
print(model_transfer)
if torch.cuda.is_available():
model_transfer = model_transfer.cuda()
print(summary(model_transfer, (3, 224, 224)))
Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.
Answer: Number of dog images are small (as compared to ImageNet which was used to train VGG), and are also dissimilar to the original training data. Hence, more layers at the end needs to be replaced, although just stripping the last layer was sufficient to get 80% accuracy to the flower classification transfer learning exercise.
Use the next code cell to specify a loss function and optimizer. Save the chosen loss function as criterion_transfer, and the optimizer as optimizer_transfer below.
criterion_transfer = nn.CrossEntropyLoss()
optimizer_transfer = optim.SGD(model_transfer.classifier[6].parameters(), lr=0.001)
Train and validate your model in the code cell below. Save the final model parameters at filepath 'model_transfer.pt'.
# train the model
import numpy as np
n_epochs = 5
model_transfer = train(n_epochs, loaders_transfer, model_transfer, optimizer_transfer, criterion_transfer, use_cuda, 'model_transfer.pt')
# load the model that got the best validation accuracy (uncomment the line below)
#model_transfer.load_state_dict(torch.load('model_transfer.pt'))
test(loaders_transfer, model_transfer, criterion_transfer, use_cuda)
model_transfer = train(n_epochs, loaders_transfer, model_transfer, optimizer_transfer, criterion_transfer, use_cuda, 'model_transfer1.pt')
test(loaders_transfer, model_transfer, criterion_transfer, use_cuda)
Results (VGG16):
Epoch: 1 Training Loss: 4.047006 Validation Loss: 3.030503
Validation loss decreased (inf --> 3.030503). Saving model ...
Epoch: 2 Training Loss: 2.575264 Validation Loss: 2.067370
Validation loss decreased (3.030503 --> 2.067370). Saving model ...
Epoch: 3 Training Loss: 1.867757 Validation Loss: 1.594315
Validation loss decreased (2.067370 --> 1.594315). Saving model ...
Epoch: 4 Training Loss: 1.495863 Validation Loss: 1.318816
Validation loss decreased (1.594315 --> 1.318816). Saving model ...
Epoch: 5 Training Loss: 1.272060 Validation Loss: 1.142953
Validation loss decreased (1.318816 --> 1.142953). Saving model ...
Results (ResNet18):
Epoch: 1 Training Loss: 12.559216 Validation Loss: 46.979374
Epoch: 1 Training Loss: ~5 Validation Loss: ~<5
Validation loss increase
Epoch: 1 Training Loss: 4.687735 Validation Loss: 5.289638
Validation loss decreased (inf --> 5.289638). Saving model ...
Epoch: 2 Training Loss: 4.770319 Validation Loss: 5.207141
Validation loss decreased (5.289638 --> 5.207141). Saving model ...
Epoch: 3 Training Loss: 4.753299 Validation Loss: 5.078537
Validation loss decreased (5.207141 --> 5.078537). Saving model ...
Epoch: 4 Training Loss: 4.726197 Validation Loss: 4.965718
Validation loss decreased (5.078537 --> 4.965718). Saving model ...
Epoch: 5 Training Loss: 4.701176 Validation Loss: 4.846803
Validation loss decreased (4.965718 --> 4.846803). Saving model ...
Epoch: 6 Training Loss: 4.678950 Validation Loss: 4.750106
Validation loss decreased (4.846803 --> 4.750106). Saving model ...
Epoch: 1 Training Loss: 5.086270 Validation Loss: 4.975640
Validation loss decreased (inf --> 4.975640). Saving model ...
Epoch: 2 Training Loss: 5.046070 Validation Loss: 4.939723
Validation loss decreased (4.975640 --> 4.939723). Saving model ...
Epoch: 3 Training Loss: 5.037471 Validation Loss: 4.910023
Validation loss decreased (4.939723 --> 4.910023). Saving model ...
Replace SGD with Adam
Epoch: 1 Training Loss: 8.818595 Validation Loss: 5.739745
Validation loss decreased (inf --> 5.739745). Saving model ...
Epoch: 2 Training Loss: 7.528431 Validation Loss: 7.082764
Epoch: 3 Training Loss: 7.566106 Validation Loss: 6.694774
Epoch: 4 Training Loss: 7.529278 Validation Loss: 6.519275
Epoch: 5 Training Loss: 7.489252 Validation Loss: 6.394591
Epoch: 6 Training Loss: 7.452108 Validation Loss: 6.241271
Epoch: 7 Training Loss: 7.420009 Validation Loss: 6.126494
Epoch: 8 Training Loss: 7.392484 Validation Loss: 5.986011
Epoch: 9 Training Loss: 7.360799 Validation Loss: 5.916609
Epoch: 1 Training Loss: 5.098560 Validation Loss: 5.031323
Validation loss decreased (inf --> 5.031323). Saving model ...
Epoch: 2 Training Loss: 5.055503 Validation Loss: 4.985293
Validation loss decreased (5.031323 --> 4.985293). Saving model ...
Epoch: 3 Training Loss: 5.043603 Validation Loss: 4.951829
Validation loss decreased (4.985293 --> 4.951829). Saving model ...
Repeat with ResNet50
Epoch: 1 Training Loss: 4.994135 Validation Loss: 4.904807
Validation loss decreased (inf --> 4.904807). Saving model ...
Epoch: 2 Training Loss: 4.989738 Validation Loss: 4.884143
Validation loss decreased (4.904807 --> 4.884143). Saving model ...
Observations: VGG performed better than ResNet.
Try out your model on the test dataset of dog images. Use the code cell below to calculate and print the test loss and accuracy. Ensure that your test accuracy is greater than 60%.
Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan hound, etc) that is predicted by your model.
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.
from PIL import Image
# list of class names by index, i.e. a name can be accessed like class_names[0]
class_names = [item[4:].replace("_", " ") for item in files['train'].classes]
def predict_breed_transfer(img_path):
# load the image and return the predicted breed
image = Image.open(img_path)
in_transform = transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5),
(0.5, 0.5, 0.5))])
image = in_transform(image)
# add batch dimension
image = image.unsqueeze_(0)
model_transfer.eval()
output = model_transfer(image)
index = output.data.numpy().argmax()
return class_names[index] # predicted class index
model_transfer.load_state_dict(torch.load('model_transfer1.pt', map_location=lambda storage, loc: storage))
num = np.random.random_integers(0, len(dog_files), size=5)
for i in num:
print(predict_breed_transfer(dog_files[i]))
print(dog_files[i])
Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,
You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and human_detector functions developed above. You are required to use your CNN from Step 4 to predict dog breed.
Some sample output for our algorithm is provided below, but feel free to design your own user experience!

### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.
def run_app(img_path):
## handle cases for a human face, dog, and neither
img = Image.open(img_path)
plt.imshow(img)
plt.show()
dog = dog_detector(img_path)
human = face_detector(img_path)
if dog:
print("Dog detected, predicted breed is: "+ predict_breed_transfer(img_path))
elif human:
print("Human detected, resembling dog breed is: "+ predict_breed_transfer(img_path))
else:
print("Error, neither human nor dog is detected.")
In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?
Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.
Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.
Answer: (Three possible points for improvement)
Worse than expected as it was unable to detect the human faces.
Possible points of improvements:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.
## suggested code, below
#for file in np.hstack(test_files):
# run_app(file)
for img_file in os.listdir('./Test'):
img_path = os.path.join('./Test', img_file)
#print(img_path)
if img_file[-3:]=='jpg':
run_app(img_path)
else:
print('')
import numpy as np
from glob import glob
# load filenames for human and dog images
human_files = np.array(glob("/data/lfw/*/*"))
dog_files = np.array(glob("/data/dog_images/*/*/*"))
num = np.random.random_integers(0, len(human_files), size=5)
for i in num:
run_app(human_files[i])
num = np.random.random_integers(0, len(dog_files), size=5)
for i in num:
run_app(dog_files[i])
print(dog_files[i])